a functional approach to emotion in autonomous systems

A Functional Approach to Emotion inAutonomous Systems

Ricardo Sanz, Carlos Hernandez, Jaime Gomez and Adolfo Hernando

Abstract The construction of fully effective systems seems to pass through theproper exploitation of goal-centric self-evaluative capabilities that let the systemteleologically self-manage. Emotions seem to provide this kind of functionality tobiological systems and hence the interest in emotion for function sustainment inartificial systems performing in changing and uncertain environments; far beyondthe media hullabaloo of displaying human-like emotion-laden faces in robots. Thischapter provides a brief analysis of the scientific theories of emotion and presentsan engineering approach for developing technology for robust autonomy by imple-menting functionality inspired in that of biological emotions

1 Introduction

The central tenet of engineering research is the development of technology forachieving some desired level of performance in artificial systems: 220 volt in a wallsocket, 240 k/m in a car, 80 Hz beat in a pacemaker, etc.

Once the development of the base technology – electrical engineering, mechani-cal engineering, embedded electronics – lets reach this performance level, a secondaspect gains in importance: maintaining this performance. The maintenance of a cer-tain level of performance is obvious in the pacemaker or wall socket, where pumpingrate and voltage must be maintained at certain values; it may be less obvious in thecar with its continuously varying road conditions, but is clear for the speed whichin modern vehicles is to be maintained by the cruise control system. The need forperformance keeping may be not so easy to pinpoint; e.g. for the brakes, where itis not so clear what is the variable to be kept within a certain range, although, ifcarefully analysed, you may find one: the braking power.

All authors are affiliated to the Autonomous Systems Laboratory (ASLab-UPM), UniversidadPolitecnica de Madrid, Jose Gutierrez Abascal 2 28045 Madrid, e-mail: [email protected],[email protected], [email protected] and [email protected]

1

2 Ricardo Sanz, Carlos Hernandez, Jaime Gomez and Adolfo Hernando

Fig. 1 The IST ICEA projectis focused in the extractionof integration patterns amongthe cognitive, emotional andautonomic systems of therat. These are evaluated ontechnical systems includingphysical and simulated rats.

1.1 Sustainable performance

The preservation of performance levels shall be understood in relation with the con-cepts of resilience and ultimately robust autonomy, which relates to the capability ofsystems to keep their proper functioning despite the uncertain and sometimes ham-pering and even hazardous dynamics of the environment where they perform. Thisidea of sustaining performance obviously maps in different ways to different kindsof applications that have more or less resilient structures in changing environments.

In our research of robust autonomy, we focus on two domains of artificial sys-tems: large-scale industrial plants and mobile robotics. While in the case of indus-trial plants sustaining performance maps crystal clear to maintaining the productionrate and keep it within a quality range, this mapping in a mobile robot scenario isfar from evident.

In order to achieve robust autonomy, one approach could be to build systemsphysically robust, so as to minimise the effect of external disturbances from theenvironment. However this approach is always cost prohibitive – i.e. let us thinkof building a bulldozer-like field robot so that it does not need to avoid obstacles –not to say usually unrealisable – consider a thermodynamically isolated kiln withno need for a temperature control. Discarded this somewhat outdated ‘brute force’strategy, engineers now must look for a more “intelligent” approach, where such aterm does not only refers to smartness by their part, but also to the capability of thebuilt systems to take advantage of information [26].

To effectively and efficiently address the problem of robust autonomy, the mate-rial and energy flows from/to the environment ought to be accompanied by the cor-responding flows of information that enable the systems to manage the uncertaintyin the operational conditions. The artificial systems must build informational struc-tures which implement relevant information about the environment, themselves andthe interaction between both. Consequently, the systems are operationalised; theycan cope successfully with the inherently dynamic environment, maximising theireffectivity when they pursue the performance goal they were designed for. The con-struction of maximally effective systems seems therefore to pass through the proper

A Functional Approach to Emotion in Autonomous Systems 3

exploitation of goal-centric self-evaluative capabilities that let the systems teleolog-ically self-manage.

In the context of the EU-funded project ICEA1, this search for resilience is fo-cused on the way that cognitive, emotional and autonomic subsystems are integratedinto a single control architecture: that of the mammal brain. The special focus onemotions is due to the fact that emotions, conceived here as internal states and pro-cesses, seem to play this kind of role of valence-centric modification in biologicalsystems. Several brain subsystems are being investigated in this direction, especiallybasal ganglia, amygdala and hippocampus, because of their involvement in basicemotional and cognitive processes: the hippocampus plays a key role in spatial cog-nition and memory, the amygdala is a main centre for emotion and basal ganglia areinvolved in valenced decision making.

1.2 Emotion for engineering sustainable performance

Most research in applying emotion to technical systems has focused on the involve-ment of the display of emotions[11] in social interaction and communication, and itsapplication for the improvement of human-artificial systems interfaces (see Figure2).

However, beyond the obvious capability for displaying human-like, emotion-laden faces in robots (see Figure 3), emotional mechanisms may play a criticalrole in the sustainment of function in changing environments. From a laymans un-derstanding of this hypothesis, we could take the example of the fear one wouldexperience upon discovering an intruder in ones home. That strong emotion elicitsa different state in humans: we enhance our attention to sensory systems (i.e. hear-ing), the discharge of adrenaline prepares our motor system for faster responses,

Fig. 2 iCat is the Philips re-search platform for studyinghuman-robot interaction top-ics, intended to stimulate re-search in this area by buildinga research community throughsupporting a common hard-ware and software platform(from www.phillips.com).

1 www.iceaproject.eu


increases our heart rate, etc. Everything is aimed towards the maintenance of ahigh-level function which we could label as “survival”. From a more formal, sci-entific perspective, it is commonly agreed that biological emotions, as consideredin the previous example, are an evolved mechanism for adaptation related to a cer-tain appraisal of internal and external events [6]. This appraisal seems to assign acertain value to stimuli, external or internal, related to the goal tree of the system(i.e. survival, reproduce, search food, avoid predators) and helps reconfigure the sys-tem accordingly. Let us take a National Geographic example. When in the matingepoch, a gazelle may feel angry. That feeling (a bodily emotion) helps reconfiguringits behaviour toward a goal that has gain priority. Going further with this example,suppose that, when searching for food, the gazelle hears the possible presence ofa lion or other predator: now the emotion of fear enhances her sensory system, toconfirm the presence of the predator and locate it, and shifts her brain into survivalmode after the decision-reconfiguration typically called fight or flight.

This describe core functionality of emotions in biological systems parallels thefunctional needs in artificial systems – presented in the previous section – in thepursue of robustness and sustainable performance. In the rest of this chapter we willanalyse some of the most relevant theories on emotions and coalesce them into atheoretical framework – the ASys Framework – for addressing the technical issuesof building similar mechanisms into artificial systems to achieve greater levels ofresilience and performance.

2 A Review of Emotion

A common, somewhat folk, theory of human emotions say that they are, to a largeextent, subjective and non deterministic. This accounts for the obvious fact that ap-parently identical stimuli may raise different emotions in different humans, and thesame individual may experiment different emotions in response to the similar stimu-

Fig. 3 The MIT robot Kismetis a paradigmatic exampleof research in the display ofemotions by robots [7].


lus. Obviously, and from a purely systemic perspective, there must be some disparitybetween systems if the behaviour differs, but the ascription of the variety in emo-tional response by dilettantes have gone from subtle details in otherwise apparentlyidentical stimulus to the very possibility of human freedom and non-determinism.

Starting at the last decades of 19th century, with the work of William James [18],and through the 20th century, the scientific community has managed to turn theancestral and pre-scientific ideas about emotions into theory-laden models, wherethe sustaining theory may be formal or not. Nowadays emotions are studied just asanother natural phenomenon of living systems, such as digestion, homeostasis; and,despite the diversity of approaches and theories, there is a consensus that emotionsare an evolved adaptivity mechanism related to situation assessment and decisionmaking [22].

Some researchers [30] have also pointed out a division regarding the scientificanalysis of emotion which comprises an external, social perspective of emotion asit is involved in communication between individuals through the display by themof emotional states and attitudes, and on the other hand a internal point of viewconsidering how emotion is involved in decision making processes.

We can summarise the usual understanding of emotions into several related as-pects [5]:

A1 - how emotional behaviour is triggered by event surrounding the agent;A2 - how emotion is manifested (displayed) by/within the agent; andA3 - how emotion is felt by the agent.

This heterogeneity notwithstanding, it is necessary to think about basic physio-logical principles going down to neural-hormonal mechanisms that make a particu-lar event “emotional”. Through this section we will summarily analyse the scientificprinciples and theories about emotions abstracted so far.

2.1 Classical models

The classical theories of James-Lange or Cannon-Bard address the causality of therelation between A1-A2-A3. The model of James-Lange states that in animals, thetriggering (A1) are experiences in the world, the autonomic nervous system thencreates physiological events such as increased heart rate or muscular tension (A2),and then emotions come as conscious feelings which come about as a result of thesephysiological changes (rather than being their cause as the Cannon-Bard model pos-tulates).

Plenty of models can be found in the literature among which we would like todistinguish – for their closeness to the ASys emotion model – the model of [2],due to its relation with the shaping of action tendencies; the model of Frijda [15],due to its perspective of emotions constituting forms of action readiness; and themodels of Plutchik [23] and James [18] in relation with the bodily basis of adaptivemechanisms.


Fig. 4 Damasio proposesa hierarchical process foremotion-raising distinguish-ing between basic autonomicmechanisms, emotion, feelingand conscious awareness ofemotion.

Body

Sensory systems

Actuator systems

Basic life regulation

Emotions

Feelings

High reason

consciousunconscious

2.2 Damasio’s Model

When relating emotion to the rest of the evolved adaptive mechanisms in humans,from the hormonal system to the more cognitive ones such as consciousness, one ofthe most complete models is that of Damasio. The somatic marker hypothesis [9]and related machinery can be used to provide a deeper understanding of emotionalsystem organisation (see Figure 4), and accounts for all A1-A2-A3, with special rel-evance for the last two aspects. This structure is among the architectures explored inthe beforehand mentioned ICEA project – www.iceaproject.eu – in order to providea coherent picture of the integration of cognitive, emotional and autonomic aspectsin mammals.

Damasio’s is a concrete proposal in line with what is needed from a scientificand technical approach to emotion and cognition [21]. Architectural approaches gobeyond the three behavioural and phenomenal aspects mentioned before (triggering,display and feeling) addressing what are the physiological mechanics for all thisfunctioning.


3 Assessing Models

Most data concerning emotion comes from experimentation on animals and humans.Data coming from these sources are widely heterogeneous; from single neuron firingpatterns, columnar behaviours or activation levels of a whole brain area to hormoneconcentrations or verbally reported psychological data. This heterogeneity in thelevel of resolution and abstraction of the data is surely a factor for the difficultyof building up a unified theory of emotion which could address such a broad vari-ety of data. For example, the neurophysiological and hormonal response to fear inmammals is quite well known [13], together with the behavioural response of rats inexperiments of fear conditioning [17]. However, there still remains missing a unifiedtheory covering the gap between the low level neural and hormonal mechanisms andthe behavioural responses elicited.

Some theoretical models of emotion and associated computational implementa-tions are being explored as a promising tool for integrative understanding of thisemotive-cognitive mechanisms. The main value this approach offers is the possibil-ity of having a precise, more rigourous methodology to grasp the core concepts andarchitecture.

In this sense we can cite Botelho [6]:

We present a preliminary definition and theory of artificial emotion viewed as a sequentialprocess comprising the appraisal of the agent global state, the generation of an emotion-signal, and an emotion-response. This theory distinguishes cognitive from affective ap-praisal on an architecture-grounded basis. Affective appraisal is performed by the affectivecomponent of the architecture; cognitive appraisal is performed by its cognitive component.

Emotions affect all levels of operation in a system, from basic life regulation toconscious, cognitive processes. We use the term transversal to indicate this fact. Theconcrete way in which system operation is affected is specific to each level. Withineach of these levels, it is specific to each organ, component and process.

In other words, this means that emotions provide a common control broadcastinfrastructure which may be used differently by each of the processes in the system.In natural systems, emotions may be conveyed by neural firings and hormones (i.e.the bodily signal broadcasting mechanisms). These mechanisms must be shared bymany organs and processes in the system, which will interpret signals accordingto their purposes and architectures. For instance, a cognitive process may interprethormonal levels to obtain auxiliary information for making a decision regardingwhat the system must do next. The same hormones may be interpreted concurrentlyby other processes in order to detect danger, risk, or a need to obtain food, forexample.

This global and multi-level character of emotions explains some distinctions ofemotion-relative phenomena present in the literature, such as Damasio’s [9, 10]:

• state of emotion,• state of feeling an emotion,• state of a feeling of an emotion made conscious.


One particular way in which emotions are transversal is by broadcasting a sum-marised picture of the system state to many of its components and processes. Thismeans not only a summary of how its components find themselves, but also a certainsense of affordance of the current scenario relative to the current system situation,processes and objectives.

This is useful to the system in order to adapt to its scenario of operation, mainlyfor three reasons:

• Emotions are fast, and are available before other more cognitive information.• Emotions, being to some extent global, contribute to co-ordination and focus

of large quantities of system processes and components, which is a factor forpreserving system cohesion [20].

• Emotions can be externalised and hence used for behavioural organisation (co-operation and competition are examples) in multi agent environments (societalbehaviour being the clearest example).

3.1 The ‘Emotional’ Facial Mimicking

This last aspect, that of externalisation of emotional states, has rendered emotionalexpression one of the main topics of emotion research (Figure 5) [11, 12].

There have been plenty of efforts made towards the implementation of emotionin machines as inspired by biosystems [29] but in most of the cases they have ne-glected addressing the core issues and have instead just focused on mimicking shal-low, observable manifestations of emotion (e.g. making robot faces a la Ekman).This work is mostly irrelevant as it does not address a sound theory behind it. Allthat is expected is some improvement in social capability by facial displaying for

Fig. 5 A big amount ofresearch on emotion has beenfocused on the expression offacial emotion neglecting theinner functional aspects of it.


human emotions. This is hopeless because the functional value of the display of anemotional state in a social interaction is based on the activation in the receptor ofthe display of a behavioural model of the displayer so as to maximise effectivenessof interaction – it is more a question of better exploiting the human capabilities thanof concrete robot competencies.

Mimicking faces is thus useless unless the operational state of the displayer iswhat is captured in the model going to be activated in the receptor. Clearly this isnot the case of human vs. robot architectures (i.e. the mental model of the receptorwill be a model of a human whereas the robot is not a human at all from an archi-tectural nor functional point of view). This issue has been widely addressed in thefield of human-computer interaction and the mental models community [16]. Whatis important then is the raising of mental states in the receptor (i.e. the activation ofmental models) that are relevant for the interaction. This can only be done if emo-tion is tightly tied to the inner operational mechanisms of the agent displaying theemotion [8].

To conclude with this subject, facial expression of emotion is an externalisationof an emotional state to help reconfigure a multi-agent organisation taking into ac-count individual agent operational states.

3.2 What emotions are and are not: ideas to solve the puzzle

After this succinct analysis, the puzzle of emotion, from an engineering point ofview, can be reduced to three core aspects:

Three questions:

• What is the function that emotional mechanisms do play?

• What is the general form of an emotional mechanism?

• What is the best strategy for emotion implementation?

The analysis of the several models of emotion produce some conclusions regard-ing what emotions are not:

• Emotions are not just sophisticated input handling, i.e. not just reacting to bears.• Emotions are not just sophisticated action generation for social affective be-

haviour, i.e. not just showing embarrassment.• Emotions are not just mechanisms for re-goaling, i.e. not just deciding to change

from eating to doing sex.

A deeper analysis abstracting from the biological mechanics into the functionalstructure renders some conclusions about how emotions work:


• Emotions do generate synthetic compact states (performing state space reduc-tion) for the effective tuning and use of evolutionary meta-controllers.

• Emotions do change the control structures at the component functional level (pat-terns and roles) of subsystems.

• Emotions operate in a global controller configuration approach rendering atransversal structural feedback architecture.

In the following section we will develop these three core ideas about emotionsin the context of a technical framework intended for the engineering of maximallyautonomous systems by applying bioinspired functional concepts.

4 The ASys Framework

The ASLab ASys Project is a long-term research project focused in the develop-ment of technology for the construction of autonomous systems. What makes ASysdifferent from other projects in this field is the extremely ambitious objective ofaddressing all the domain of autonomy. We capture this purpose in the motto “en-gineering any-x autonomous systems”. The ASys Framework is both a theoreticalframework for understanding all the relevant issues and a software-intensive tech-nological framework that enables the technically sound creation of autonomous sys-tems, where autonomy is understood in its broadest sense and not in the severelyrestricted sense of the term autonomous intelligent systems that is usually equatedto mobile reactive robots.

One of the central topics in the ASys Framework is the pervasive model-basedapproach. A truly autonomous system will be continuously using models to performits activity. An ASys system will be built using models of it. An ASys can exploitits own very models for driving its behaviour. Model-based engineering and model-based behaviour then merge into a single phenomenon: model-based autonomy. Weequate this conceptualisation with cognition.

The ASys Framework hence establishes that a system is said to be cognitive if itexploits models of other systems in its interaction with them. Models and knowledgeare then equated and the ASys Framework provides a link between the ontologicaland epistemological aspects of mind.

On the technical side, the ASys Framework follows a principled approach to au-tonomous system mind construction, the cognition as model-based behaviour beingthe first principle, so as to ground a systematic engineering approach that shall endin rendering machine consciousness [25].

These principles establish guidelines for the systematic, formally grounded de-velopment of a real-time control framework based on the control and software prin-ciples of the Integrated Control Architecture [27]. This will render a methodology,a toolset and an execution framework for the engineering of robust autonomoussystems based on the implementation of cognitive mechanisms up to the level ofconsciousness [24].


Fig. 6 The mind as model-based controller vision is acentral concept in the ASysFramework. It helps bind thedynamics of the mental (Psy)to the dynamics of the phys-ical (Phy). This should notbe understood as a dualisticposition, it is not. The minditself is part of the phenom-ena of the physical [19] andwe just focus on two aspectsof this physicality, stressingthe informational nature ofminds.

Psy

Phy

M

M-1

mapmap

sense act

^

4.1 Emotion, Consciousness and Control in ASys

The mechanisms of emotion impinge on the behavioural capability of the agent soas to prepare it for future action. This makes emotion a core capability for sophis-ticated self-management control architecture where outer control loops (emotion)determine the functioning of inner control loops (homeostasis) so as to maximisesurvivability. Damasio’s model on consciousness lays out another control loop atopof these two (see Figure 4) rendering a high-level reasoning capability. Emotionsrealise meta-controllers.

As was the case with emotions, there are plenty of models of consciousness thattry to address the relation of physiology and the three core aspects of consciousness:world-awareness, self-awareness and qualia. We can distinguish as maximally rele-vant for our work, due to their abstract, general nature, the global workspace modelof Baars [4] and the information integration model of Tononi [28].

The integrated control model of consciousness [25], also part of the ASys Frame-work, is based on the provision of self-awareness by means of model-based percep-tual mechanics.

The ASys perspective on cognition/emotion goes beyond Damasio’s approach ofputting emotions/feelings as additional layers in hierarchical controllers. Emotion isno longer another layer in the architecture but a transversal mechanism that crossesacross all layers. This is indeed a well known fact in the studies of emotion. Emo-tions do appear from the subconscious plane to the conscious surface, affecting alllevels in the cognitive structure, from the physiological up to the cognitive, social,self-conscious level.

This implies (see Figure 7) that emotional mechanics are part of each level of thecontrol hierarchy. The level of focus of the analysis is what determines the labellingused for this mechanism: basic emotion, emotion, feeling, conscious feeling, etc.In the language of information technology we would say that emotion is an aspect


Body

Sensory systems

Actuator systems

High reason

consciousunconscious

Basic life regulation

Complex life regulation

Even more complex life regulation

Emotions

Feelings

Basic emotions

Conscious Feelings

Homeostatic

Allostatic

Fig. 7 Damasio’s layering of emotion appears as labelling of the transversal emo-tional mechanisms across a layered architecture for control.

–in the computer science interpretation [14]–of the different systems that constitutethe body and mind of an autonomous agent. From a functional perspective we canalso observe that the goals pursued by such control structures go from the purelyhomeostatic mechanisms for life survival to the higher-level, socially-originated al-lostatic mechanisms for social behaviour. From hunger to embarrassment, emotionsdo share the meta-control capabilities over basic behavioural structures.

Artificial implementations of emotions are not developed yet to the same degreeas the natural. However, large, distributed, fault-tolerant systems include mecha-nisms which already play a similar role [3]. Fault detection, damage confinement,error recovery and fault treatment are based on broadcast messages and other mech-anisms shared and used by system components in analogous ways to the naturalcounterparts.


4.2 Emotion mechanics in ASys

The ASys Framework for autonomous systems is based on an architecture forsoftware-intensive, distributed, real-time control called the Integrated Control Ar-chitecture (ICa) [27]. This architecture is based on the implementation of patternsof activity across sets of distributed real-time agents. These patterns respond to theneeds of the control task that can follow a multilayered, multi-objective controlstrategy [1].

The implementation of a controller over ICa renders a collection of interactingsoftware components that realise patterns of activity as sequences of service re-quests. The software component model is of extreme importance in the implemen-tation of such controllers because it provides a common modelling framework forboth the physical components of the system under control, which are but the or-gans in a biological system and constitute the fleshly infrastructure, and the mentalcomponents, which constitute the control superstructure. Figure 8 shows three suchcomponents in a simple, layered control structure: an organ, a controller and a meta-controller.

An emotional system incorporated to ICa, according to the previous conceptuali-sation of emotion in the ASys Framework, would provide the structural mechanismsfor control pattern adaptation to the current state of affairs. Examples of this kind ofarchitecture are already available in the world of control systems (for example thepreviously mentioned fault-tolerant controllers and sliding mode controllers).

The primary effect of the emotional system is the change in the functional organ-isation of the control system of the body. In Figure 8 the emotional system changesthe functional organisation from time t to t +∆ t, concretely in this example the out-put of the meta-controller to the controller – e.g. the goal or reference, in controljargon. Both the emotional observation and control are done in terms of a valuesystems for the agent. This happens in a multi-scale, multilayer organisation thatconstitutes the integrated global controller of the agent.

Now we can provide the ASys Framework answers to the three core aspects ofemotion mentioned before:

Three answers:

• What is the general form of an emotional mechanism?A self-reorganising meta-controller.

• What is the function that emotional mechanisms do play?Provide value-centric functional reorganisation.

• What is the best strategy for emotion implementation?Functional modularisation of control functions and integration over acommon infrastructure.


Emotional System

uncontrolledinputs side effects

controls controlledoutputs

Organ

Controllerconstraints

disturbances

MetaControl

EmotionalController

side effects

side effects

controls controlledoutputs

Organ

Controller

constraints

disturbances

side effects

uncontrolledinputs

EmotionalObserver(Sensor)

emotional state

entailment change

t + ∆t

MetaControl

t

Fig. 8 The figure depicts how an emotional system following these ideas would performreconfiguration in an modularised control architecture. The emotional system changes thefunctional organisation to adapt it to new operating conditions.

5 Summary

The chapter has reviewed some of the common approaches to emotion understand-ing, with an special emphasis on Damasio’s model of emotion and feeling.

It has also been analysed the extended functional role that emotions can play incomplex adaptive controllers and how the different aspects of emotion (triggering,emotional states, bodily effect) are addressed from this perspective.

This understanding has been put in the context of the ASys Framework, a the-oretical and technical framework for the implementation of autonomous systems.This framework is based on the construction of modular, component-based controlsystems following the architectural guidelines of the Integrated Control Architecture(ICa) – a software architecture based on distributed real-time objects.

This framework is being applied to the modelling and understanding of autonomic-emotional-cognitive integration aspects in the rat brain and the implementation ofembedded controllers in the context of the IST ICEA project.

Acknowledgements Authors would like to acknowledge the support coming from the EuropeanCommunity’s Seventh Framework Programme FP6/2004-2007 under grant agreement IST 027819ICEA – Integrating Cognition, Emotion and Autonomy and the Spanish Plan Nacional de I+Dunder grant agreement DPI-2006-11798 C3 – Control Cognitivo Consciente.


References

1. Alarcon, I., Rodriguez-Marin, P., Almeida, L., Sanz, R., Fontaine, L., Gomez, P., Alaman,X., Nordin, P., Bejder, H., de Pablo, E.: Heterogeneous integration architecture for intelligentcontrol systems. Intelligent Systems Engineering 3(3), 138–152 (1994)

2. Arnold, M.: Emotions and Personality. Cambridge University Press (1960)3. Astrom, K., Albertos, P., Blanke, M., Isidori, A., Schaufelberger, W., Sanz, R. (eds.): Control

of Complex Systems. Springer (2001)4. Baars, B.J.: In the theatre of consciousness. global workspace theory, a rigorous scientific

theory of consciousness. Journal of Consciousness Studies 4, 292–309 (1997)5. Bermejo Alonso, J.: A state of the art on emotion. Tech. Rep. R-2006-002, UPM Autonomous

Systems Laboratory, Universidad Politecnica de Madrid (2006)6. Botelho, L.: Machinery for artificial emotions. Cybernetics and Systems: An International

Journal 32, 465–506 (2001)7. Breazeal, C.: Sociable machines: Expressive social exchange between humans and robots.

Ph.D. thesis, Department of Electrical Engineering and Computer Science, MIT (2000)8. Conde, R.P.: Mapeo facial de emociones sinteticas. Master’s thesis, Universidad Politecnica

de Madrid (2005)9. Damasio, A.: The feeling of what happens: body and emotion in the making of consciousness.

Harcourt Press, New York (1999)10. Damasio, A.: Emotions and feelings: a neurological perspective. In: A. Manstead, N. Fridja,

A. Fischer (eds.) Feelings and emotions, chap. 4, pp. 49–57. Cambridge University Press(2004)

11. Darwin, C. (ed.): The expression of the emotions in man and animals. D. Appleton andCompany (1872)

12. Ekman, P. (ed.): Emotion in the human face. Cambridge University Press (1982)13. Fendt, M., Fanselow, M.S.: The neuroanatomical and neurochemical basis of conditioned fear.

Neuroscience & Biobehavioral Reviews 23(5), 743–760 (1999)14. Filman, R.E., Elrad, T., Clarke, S., Aksit, M. (eds.): Aspect Oriented Software Development.

Addison Wesley (2004)15. Frijda, N.: The emotions. Cambridge University Press (1987)16. Gentner, D., Stevens, A.L. (eds.): Mental models. Lawrence Erlbaum Associates, Hillsdale,

NJ (1983)17. Hatfield, T., Han, J.S., Conley, M., Gallagher, M., Holland, P.: Neurotoxic lesions of basolat-

eral, but not central, amygdala interfere with pavlovian second-order conditioning and rein-forcer devaluation effects. J. Neurosci. 16(16), 5256–5265 (1996)

18. James, W.: What is an emotion’. Mind 9, 188–205 (1884)19. Landauer, R.: Information is physical. In: Workshop on Physics and Computation, PhysComp

’92., pp. 1–4 (1992)20. Lopez, I.: A framework for perception in autonomous systems. Ph.D. thesis, Departamento

de Automatica, Universidad Politecnica de Madrid (2007)21. Ortony, A., Clore, G.L., Collins, A.: The Cognitive Structure of Emotions. Cambridge Uni-

versity Press (1988)22. Panksepp, J.: Affective neuroscience: the foundations of human and an-

imal emotions. Oxford University Press, New York (1998). URLhttp://www.loc.gov/catdir/enhancements/fy0635/98015955-d.html

23. Plutchik, R., Kellerman, H. (eds.): Emotion: Theory, research and experience. Volume I: The-ories of emotion. Academic Press, New York (1980)

24. Sanz, R., Lopez, I., Bermejo-Alonso, J., Chinchilla, R., Conde, R.: Self-X: The control within.In: Proceedings of IFAC World Congress 2005 (2005)

25. Sanz, R., Lopez, I., Rodrıguez, M., Hernandez, C.: Principles for consciousness inintegrated cognitive control. Neural Networks 20(9), 938–946 (2007). DOIhttp://dx.doi.org/10.1016/j.neunet.2007.09.012


26. Sanz, R., Matıa, F., Galan, S.: Fridges, elephants and the meaning of autonomy and intelli-gence. In: IEEE International Symposium on Intelligent Control, ISIC’2000. Patras, Greece(2000)

27. Sanz, R., Matıa, F., Puente, E.A.: The ICa approach to intelligent autonomous systems. In:S. Tzafestas (ed.) Advances in Autonomous Intelligent Systems, Microprocessor-Based andIntelligent Systems Engineering, chap. 4, pp. 71–92. Kluwer Academic Publishers, Dordretch,NL (1999)

28. Tononi, G.: An information integration theory of consciousness. BMC Neuroscience 5, 42(2004)

29. Trapl, R., Petta, P., Payr, S. (eds.): Emotion in Humans and Artifacts. MIT Press (2002)30. Ventura, R., Pinto-Ferreira, C.: Emotion–based agents: three approaches to implmentation. In:

J.D. Velasquez (ed.) Workshop on Emotion–based agent architectures. Seattle, U.S.A. (1999)

a functional approach to emotion in autonomous systems

Documents